1 // SPDX-License-Identifier: GPL-2.0
3 * PCI Message Signaled Interrupt (MSI)
5 * Copyright (C) 2003-2004 Intel
7 * Copyright (C) 2016 Christoph Hellwig.
10 #include <linux/err.h>
12 #include <linux/irq.h>
13 #include <linux/interrupt.h>
14 #include <linux/export.h>
15 #include <linux/ioport.h>
16 #include <linux/pci.h>
17 #include <linux/proc_fs.h>
18 #include <linux/msi.h>
19 #include <linux/smp.h>
20 #include <linux/errno.h>
22 #include <linux/acpi_iort.h>
23 #include <linux/slab.h>
24 #include <linux/irqdomain.h>
25 #include <linux/of_irq.h>
29 static int pci_msi_enable = 1;
30 int pci_msi_ignore_mask;
32 #define msix_table_size(flags) ((flags & PCI_MSIX_FLAGS_QSIZE) + 1)
34 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
35 static int pci_msi_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
37 struct irq_domain *domain;
39 domain = dev_get_msi_domain(&dev->dev);
40 if (domain && irq_domain_is_hierarchy(domain))
41 return msi_domain_alloc_irqs(domain, &dev->dev, nvec);
43 return arch_setup_msi_irqs(dev, nvec, type);
46 static void pci_msi_teardown_msi_irqs(struct pci_dev *dev)
48 struct irq_domain *domain;
50 domain = dev_get_msi_domain(&dev->dev);
51 if (domain && irq_domain_is_hierarchy(domain))
52 msi_domain_free_irqs(domain, &dev->dev);
54 arch_teardown_msi_irqs(dev);
57 #define pci_msi_setup_msi_irqs arch_setup_msi_irqs
58 #define pci_msi_teardown_msi_irqs arch_teardown_msi_irqs
63 int __weak arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
65 struct msi_controller *chip = dev->bus->msi;
68 if (!chip || !chip->setup_irq)
71 err = chip->setup_irq(chip, dev, desc);
75 irq_set_chip_data(desc->irq, chip);
80 void __weak arch_teardown_msi_irq(unsigned int irq)
82 struct msi_controller *chip = irq_get_chip_data(irq);
84 if (!chip || !chip->teardown_irq)
87 chip->teardown_irq(chip, irq);
90 int __weak arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
92 struct msi_controller *chip = dev->bus->msi;
93 struct msi_desc *entry;
96 if (chip && chip->setup_irqs)
97 return chip->setup_irqs(chip, dev, nvec, type);
99 * If an architecture wants to support multiple MSI, it needs to
100 * override arch_setup_msi_irqs()
102 if (type == PCI_CAP_ID_MSI && nvec > 1)
105 for_each_pci_msi_entry(entry, dev) {
106 ret = arch_setup_msi_irq(dev, entry);
117 * We have a default implementation available as a separate non-weak
118 * function, as it is used by the Xen x86 PCI code
120 void default_teardown_msi_irqs(struct pci_dev *dev)
123 struct msi_desc *entry;
125 for_each_pci_msi_entry(entry, dev)
127 for (i = 0; i < entry->nvec_used; i++)
128 arch_teardown_msi_irq(entry->irq + i);
131 void __weak arch_teardown_msi_irqs(struct pci_dev *dev)
133 return default_teardown_msi_irqs(dev);
136 static void default_restore_msi_irq(struct pci_dev *dev, int irq)
138 struct msi_desc *entry;
141 if (dev->msix_enabled) {
142 for_each_pci_msi_entry(entry, dev) {
143 if (irq == entry->irq)
146 } else if (dev->msi_enabled) {
147 entry = irq_get_msi_desc(irq);
151 __pci_write_msi_msg(entry, &entry->msg);
154 void __weak arch_restore_msi_irqs(struct pci_dev *dev)
156 return default_restore_msi_irqs(dev);
159 static inline __attribute_const__ u32 msi_mask(unsigned x)
161 /* Don't shift by >= width of type */
164 return (1 << (1 << x)) - 1;
168 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
169 * mask all MSI interrupts by clearing the MSI enable bit does not work
170 * reliably as devices without an INTx disable bit will then generate a
171 * level IRQ which will never be cleared.
173 u32 __pci_msi_desc_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
175 u32 mask_bits = desc->masked;
177 if (pci_msi_ignore_mask || !desc->msi_attrib.maskbit)
182 pci_write_config_dword(msi_desc_to_pci_dev(desc), desc->mask_pos,
188 static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
190 desc->masked = __pci_msi_desc_mask_irq(desc, mask, flag);
193 static void __iomem *pci_msix_desc_addr(struct msi_desc *desc)
195 if (desc->msi_attrib.is_virtual)
198 return desc->mask_base +
199 desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
203 * This internal function does not flush PCI writes to the device.
204 * All users must ensure that they read from the device before either
205 * assuming that the device state is up to date, or returning out of this
206 * file. This saves a few milliseconds when initialising devices with lots
207 * of MSI-X interrupts.
209 u32 __pci_msix_desc_mask_irq(struct msi_desc *desc, u32 flag)
211 u32 mask_bits = desc->masked;
212 void __iomem *desc_addr;
214 if (pci_msi_ignore_mask)
216 desc_addr = pci_msix_desc_addr(desc);
220 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
222 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
224 writel(mask_bits, desc_addr + PCI_MSIX_ENTRY_VECTOR_CTRL);
229 static void msix_mask_irq(struct msi_desc *desc, u32 flag)
231 desc->masked = __pci_msix_desc_mask_irq(desc, flag);
234 static void msi_set_mask_bit(struct irq_data *data, u32 flag)
236 struct msi_desc *desc = irq_data_get_msi_desc(data);
238 if (desc->msi_attrib.is_msix) {
239 msix_mask_irq(desc, flag);
240 readl(desc->mask_base); /* Flush write to device */
242 unsigned offset = data->irq - desc->irq;
243 msi_mask_irq(desc, 1 << offset, flag << offset);
248 * pci_msi_mask_irq - Generic IRQ chip callback to mask PCI/MSI interrupts
249 * @data: pointer to irqdata associated to that interrupt
251 void pci_msi_mask_irq(struct irq_data *data)
253 msi_set_mask_bit(data, 1);
255 EXPORT_SYMBOL_GPL(pci_msi_mask_irq);
258 * pci_msi_unmask_irq - Generic IRQ chip callback to unmask PCI/MSI interrupts
259 * @data: pointer to irqdata associated to that interrupt
261 void pci_msi_unmask_irq(struct irq_data *data)
263 msi_set_mask_bit(data, 0);
265 EXPORT_SYMBOL_GPL(pci_msi_unmask_irq);
267 void default_restore_msi_irqs(struct pci_dev *dev)
269 struct msi_desc *entry;
271 for_each_pci_msi_entry(entry, dev)
272 default_restore_msi_irq(dev, entry->irq);
275 void __pci_read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
277 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
279 BUG_ON(dev->current_state != PCI_D0);
281 if (entry->msi_attrib.is_msix) {
282 void __iomem *base = pci_msix_desc_addr(entry);
289 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
290 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
291 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
293 int pos = dev->msi_cap;
296 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
298 if (entry->msi_attrib.is_64) {
299 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
301 pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
304 pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
310 void __pci_write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
312 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
314 if (dev->current_state != PCI_D0 || pci_dev_is_disconnected(dev)) {
315 /* Don't touch the hardware now */
316 } else if (entry->msi_attrib.is_msix) {
317 void __iomem *base = pci_msix_desc_addr(entry);
322 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
323 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
324 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
326 int pos = dev->msi_cap;
329 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
330 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
331 msgctl |= entry->msi_attrib.multiple << 4;
332 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
334 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
336 if (entry->msi_attrib.is_64) {
337 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
339 pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
342 pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
350 if (entry->write_msi_msg)
351 entry->write_msi_msg(entry, entry->write_msi_msg_data);
355 void pci_write_msi_msg(unsigned int irq, struct msi_msg *msg)
357 struct msi_desc *entry = irq_get_msi_desc(irq);
359 __pci_write_msi_msg(entry, msg);
361 EXPORT_SYMBOL_GPL(pci_write_msi_msg);
363 static void free_msi_irqs(struct pci_dev *dev)
365 struct list_head *msi_list = dev_to_msi_list(&dev->dev);
366 struct msi_desc *entry, *tmp;
367 struct attribute **msi_attrs;
368 struct device_attribute *dev_attr;
371 for_each_pci_msi_entry(entry, dev)
373 for (i = 0; i < entry->nvec_used; i++)
374 BUG_ON(irq_has_action(entry->irq + i));
376 pci_msi_teardown_msi_irqs(dev);
378 list_for_each_entry_safe(entry, tmp, msi_list, list) {
379 if (entry->msi_attrib.is_msix) {
380 if (list_is_last(&entry->list, msi_list))
381 iounmap(entry->mask_base);
384 list_del(&entry->list);
385 free_msi_entry(entry);
388 if (dev->msi_irq_groups) {
389 sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
390 msi_attrs = dev->msi_irq_groups[0]->attrs;
391 while (msi_attrs[count]) {
392 dev_attr = container_of(msi_attrs[count],
393 struct device_attribute, attr);
394 kfree(dev_attr->attr.name);
399 kfree(dev->msi_irq_groups[0]);
400 kfree(dev->msi_irq_groups);
401 dev->msi_irq_groups = NULL;
405 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
407 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
408 pci_intx(dev, enable);
411 static void __pci_restore_msi_state(struct pci_dev *dev)
414 struct msi_desc *entry;
416 if (!dev->msi_enabled)
419 entry = irq_get_msi_desc(dev->irq);
421 pci_intx_for_msi(dev, 0);
422 pci_msi_set_enable(dev, 0);
423 arch_restore_msi_irqs(dev);
425 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
426 msi_mask_irq(entry, msi_mask(entry->msi_attrib.multi_cap),
428 control &= ~PCI_MSI_FLAGS_QSIZE;
429 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
430 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
433 static void __pci_restore_msix_state(struct pci_dev *dev)
435 struct msi_desc *entry;
437 if (!dev->msix_enabled)
439 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
441 /* route the table */
442 pci_intx_for_msi(dev, 0);
443 pci_msix_clear_and_set_ctrl(dev, 0,
444 PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
446 arch_restore_msi_irqs(dev);
447 for_each_pci_msi_entry(entry, dev)
448 msix_mask_irq(entry, entry->masked);
450 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
453 void pci_restore_msi_state(struct pci_dev *dev)
455 __pci_restore_msi_state(dev);
456 __pci_restore_msix_state(dev);
458 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
460 static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
463 struct msi_desc *entry;
467 retval = kstrtoul(attr->attr.name, 10, &irq);
471 entry = irq_get_msi_desc(irq);
473 return sprintf(buf, "%s\n",
474 entry->msi_attrib.is_msix ? "msix" : "msi");
479 static int populate_msi_sysfs(struct pci_dev *pdev)
481 struct attribute **msi_attrs;
482 struct attribute *msi_attr;
483 struct device_attribute *msi_dev_attr;
484 struct attribute_group *msi_irq_group;
485 const struct attribute_group **msi_irq_groups;
486 struct msi_desc *entry;
492 /* Determine how many msi entries we have */
493 for_each_pci_msi_entry(entry, pdev)
494 num_msi += entry->nvec_used;
498 /* Dynamically create the MSI attributes for the PCI device */
499 msi_attrs = kcalloc(num_msi + 1, sizeof(void *), GFP_KERNEL);
502 for_each_pci_msi_entry(entry, pdev) {
503 for (i = 0; i < entry->nvec_used; i++) {
504 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
507 msi_attrs[count] = &msi_dev_attr->attr;
509 sysfs_attr_init(&msi_dev_attr->attr);
510 msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
512 if (!msi_dev_attr->attr.name)
514 msi_dev_attr->attr.mode = S_IRUGO;
515 msi_dev_attr->show = msi_mode_show;
520 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
523 msi_irq_group->name = "msi_irqs";
524 msi_irq_group->attrs = msi_attrs;
526 msi_irq_groups = kcalloc(2, sizeof(void *), GFP_KERNEL);
528 goto error_irq_group;
529 msi_irq_groups[0] = msi_irq_group;
531 ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
533 goto error_irq_groups;
534 pdev->msi_irq_groups = msi_irq_groups;
539 kfree(msi_irq_groups);
541 kfree(msi_irq_group);
544 msi_attr = msi_attrs[count];
546 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
547 kfree(msi_attr->name);
550 msi_attr = msi_attrs[count];
556 static struct msi_desc *
557 msi_setup_entry(struct pci_dev *dev, int nvec, struct irq_affinity *affd)
559 struct irq_affinity_desc *masks = NULL;
560 struct msi_desc *entry;
564 masks = irq_create_affinity_masks(nvec, affd);
566 /* MSI Entry Initialization */
567 entry = alloc_msi_entry(&dev->dev, nvec, masks);
571 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
573 entry->msi_attrib.is_msix = 0;
574 entry->msi_attrib.is_64 = !!(control & PCI_MSI_FLAGS_64BIT);
575 entry->msi_attrib.is_virtual = 0;
576 entry->msi_attrib.entry_nr = 0;
577 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
578 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
579 entry->msi_attrib.multi_cap = (control & PCI_MSI_FLAGS_QMASK) >> 1;
580 entry->msi_attrib.multiple = ilog2(__roundup_pow_of_two(nvec));
582 if (control & PCI_MSI_FLAGS_64BIT)
583 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
585 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
587 /* Save the initial mask status */
588 if (entry->msi_attrib.maskbit)
589 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
596 static int msi_verify_entries(struct pci_dev *dev)
598 struct msi_desc *entry;
600 for_each_pci_msi_entry(entry, dev) {
601 if (!dev->no_64bit_msi || !entry->msg.address_hi)
603 pci_err(dev, "Device has broken 64-bit MSI but arch"
604 " tried to assign one above 4G\n");
611 * msi_capability_init - configure device's MSI capability structure
612 * @dev: pointer to the pci_dev data structure of MSI device function
613 * @nvec: number of interrupts to allocate
614 * @affd: description of automatic IRQ affinity assignments (may be %NULL)
616 * Setup the MSI capability structure of the device with the requested
617 * number of interrupts. A return value of zero indicates the successful
618 * setup of an entry with the new MSI IRQ. A negative return value indicates
619 * an error, and a positive return value indicates the number of interrupts
620 * which could have been allocated.
622 static int msi_capability_init(struct pci_dev *dev, int nvec,
623 struct irq_affinity *affd)
625 struct msi_desc *entry;
629 pci_msi_set_enable(dev, 0); /* Disable MSI during set up */
631 entry = msi_setup_entry(dev, nvec, affd);
635 /* All MSIs are unmasked by default; mask them all */
636 mask = msi_mask(entry->msi_attrib.multi_cap);
637 msi_mask_irq(entry, mask, mask);
639 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
641 /* Configure MSI capability structure */
642 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
644 msi_mask_irq(entry, mask, ~mask);
649 ret = msi_verify_entries(dev);
651 msi_mask_irq(entry, mask, ~mask);
656 ret = populate_msi_sysfs(dev);
658 msi_mask_irq(entry, mask, ~mask);
663 /* Set MSI enabled bits */
664 pci_intx_for_msi(dev, 0);
665 pci_msi_set_enable(dev, 1);
666 dev->msi_enabled = 1;
668 pcibios_free_irq(dev);
669 dev->irq = entry->irq;
673 static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
675 resource_size_t phys_addr;
680 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
682 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
683 flags = pci_resource_flags(dev, bir);
684 if (!flags || (flags & IORESOURCE_UNSET))
687 table_offset &= PCI_MSIX_TABLE_OFFSET;
688 phys_addr = pci_resource_start(dev, bir) + table_offset;
690 return ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
693 static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
694 struct msix_entry *entries, int nvec,
695 struct irq_affinity *affd)
697 struct irq_affinity_desc *curmsk, *masks = NULL;
698 struct msi_desc *entry;
700 int vec_count = pci_msix_vec_count(dev);
703 masks = irq_create_affinity_masks(nvec, affd);
705 for (i = 0, curmsk = masks; i < nvec; i++) {
706 entry = alloc_msi_entry(&dev->dev, 1, curmsk);
712 /* No enough memory. Don't try again */
717 entry->msi_attrib.is_msix = 1;
718 entry->msi_attrib.is_64 = 1;
720 entry->msi_attrib.entry_nr = entries[i].entry;
722 entry->msi_attrib.entry_nr = i;
724 entry->msi_attrib.is_virtual =
725 entry->msi_attrib.entry_nr >= vec_count;
727 entry->msi_attrib.default_irq = dev->irq;
728 entry->mask_base = base;
730 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
740 static void msix_program_entries(struct pci_dev *dev,
741 struct msix_entry *entries)
743 struct msi_desc *entry;
745 void __iomem *desc_addr;
747 for_each_pci_msi_entry(entry, dev) {
749 entries[i++].vector = entry->irq;
751 desc_addr = pci_msix_desc_addr(entry);
753 entry->masked = readl(desc_addr +
754 PCI_MSIX_ENTRY_VECTOR_CTRL);
758 msix_mask_irq(entry, 1);
763 * msix_capability_init - configure device's MSI-X capability
764 * @dev: pointer to the pci_dev data structure of MSI-X device function
765 * @entries: pointer to an array of struct msix_entry entries
766 * @nvec: number of @entries
767 * @affd: Optional pointer to enable automatic affinity assignment
769 * Setup the MSI-X capability structure of device function with a
770 * single MSI-X IRQ. A return of zero indicates the successful setup of
771 * requested MSI-X entries with allocated IRQs or non-zero for otherwise.
773 static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,
774 int nvec, struct irq_affinity *affd)
780 /* Ensure MSI-X is disabled while it is set up */
781 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
783 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
784 /* Request & Map MSI-X table region */
785 base = msix_map_region(dev, msix_table_size(control));
789 ret = msix_setup_entries(dev, base, entries, nvec, affd);
793 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
797 /* Check if all MSI entries honor device restrictions */
798 ret = msi_verify_entries(dev);
803 * Some devices require MSI-X to be enabled before we can touch the
804 * MSI-X registers. We need to mask all the vectors to prevent
805 * interrupts coming in before they're fully set up.
807 pci_msix_clear_and_set_ctrl(dev, 0,
808 PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE);
810 msix_program_entries(dev, entries);
812 ret = populate_msi_sysfs(dev);
816 /* Set MSI-X enabled bits and unmask the function */
817 pci_intx_for_msi(dev, 0);
818 dev->msix_enabled = 1;
819 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
821 pcibios_free_irq(dev);
827 * If we had some success, report the number of IRQs
828 * we succeeded in setting up.
830 struct msi_desc *entry;
833 for_each_pci_msi_entry(entry, dev) {
848 * pci_msi_supported - check whether MSI may be enabled on a device
849 * @dev: pointer to the pci_dev data structure of MSI device function
850 * @nvec: how many MSIs have been requested?
852 * Look at global flags, the device itself, and its parent buses
853 * to determine if MSI/-X are supported for the device. If MSI/-X is
854 * supported return 1, else return 0.
856 static int pci_msi_supported(struct pci_dev *dev, int nvec)
860 /* MSI must be globally enabled and supported by the device */
864 if (!dev || dev->no_msi || dev->current_state != PCI_D0)
868 * You can't ask to have 0 or less MSIs configured.
870 * b) the list manipulation code assumes nvec >= 1.
876 * Any bridge which does NOT route MSI transactions from its
877 * secondary bus to its primary bus must set NO_MSI flag on
878 * the secondary pci_bus.
879 * We expect only arch-specific PCI host bus controller driver
880 * or quirks for specific PCI bridges to be setting NO_MSI.
882 for (bus = dev->bus; bus; bus = bus->parent)
883 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
890 * pci_msi_vec_count - Return the number of MSI vectors a device can send
891 * @dev: device to report about
893 * This function returns the number of MSI vectors a device requested via
894 * Multiple Message Capable register. It returns a negative errno if the
895 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
896 * and returns a power of two, up to a maximum of 2^5 (32), according to the
899 int pci_msi_vec_count(struct pci_dev *dev)
907 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
908 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
912 EXPORT_SYMBOL(pci_msi_vec_count);
914 static void pci_msi_shutdown(struct pci_dev *dev)
916 struct msi_desc *desc;
919 if (!pci_msi_enable || !dev || !dev->msi_enabled)
922 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
923 desc = first_pci_msi_entry(dev);
925 pci_msi_set_enable(dev, 0);
926 pci_intx_for_msi(dev, 1);
927 dev->msi_enabled = 0;
929 /* Return the device with MSI unmasked as initial states */
930 mask = msi_mask(desc->msi_attrib.multi_cap);
931 /* Keep cached state to be restored */
932 __pci_msi_desc_mask_irq(desc, mask, ~mask);
934 /* Restore dev->irq to its default pin-assertion IRQ */
935 dev->irq = desc->msi_attrib.default_irq;
936 pcibios_alloc_irq(dev);
939 void pci_disable_msi(struct pci_dev *dev)
941 if (!pci_msi_enable || !dev || !dev->msi_enabled)
944 pci_msi_shutdown(dev);
947 EXPORT_SYMBOL(pci_disable_msi);
950 * pci_msix_vec_count - return the number of device's MSI-X table entries
951 * @dev: pointer to the pci_dev data structure of MSI-X device function
952 * This function returns the number of device's MSI-X table entries and
953 * therefore the number of MSI-X vectors device is capable of sending.
954 * It returns a negative errno if the device is not capable of sending MSI-X
957 int pci_msix_vec_count(struct pci_dev *dev)
964 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
965 return msix_table_size(control);
967 EXPORT_SYMBOL(pci_msix_vec_count);
969 static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
970 int nvec, struct irq_affinity *affd, int flags)
975 if (!pci_msi_supported(dev, nvec))
978 nr_entries = pci_msix_vec_count(dev);
981 if (nvec > nr_entries && !(flags & PCI_IRQ_VIRTUAL))
985 /* Check for any invalid entries */
986 for (i = 0; i < nvec; i++) {
987 if (entries[i].entry >= nr_entries)
988 return -EINVAL; /* invalid entry */
989 for (j = i + 1; j < nvec; j++) {
990 if (entries[i].entry == entries[j].entry)
991 return -EINVAL; /* duplicate entry */
996 /* Check whether driver already requested for MSI IRQ */
997 if (dev->msi_enabled) {
998 pci_info(dev, "can't enable MSI-X (MSI IRQ already assigned)\n");
1001 return msix_capability_init(dev, entries, nvec, affd);
1004 static void pci_msix_shutdown(struct pci_dev *dev)
1006 struct msi_desc *entry;
1008 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1011 if (pci_dev_is_disconnected(dev)) {
1012 dev->msix_enabled = 0;
1016 /* Return the device with MSI-X masked as initial states */
1017 for_each_pci_msi_entry(entry, dev) {
1018 /* Keep cached states to be restored */
1019 __pci_msix_desc_mask_irq(entry, 1);
1022 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
1023 pci_intx_for_msi(dev, 1);
1024 dev->msix_enabled = 0;
1025 pcibios_alloc_irq(dev);
1028 void pci_disable_msix(struct pci_dev *dev)
1030 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1033 pci_msix_shutdown(dev);
1036 EXPORT_SYMBOL(pci_disable_msix);
1038 void pci_no_msi(void)
1044 * pci_msi_enabled - is MSI enabled?
1046 * Returns true if MSI has not been disabled by the command-line option
1049 int pci_msi_enabled(void)
1051 return pci_msi_enable;
1053 EXPORT_SYMBOL(pci_msi_enabled);
1055 static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
1056 struct irq_affinity *affd)
1061 if (!pci_msi_supported(dev, minvec))
1064 /* Check whether driver already requested MSI-X IRQs */
1065 if (dev->msix_enabled) {
1066 pci_info(dev, "can't enable MSI (MSI-X already enabled)\n");
1070 if (maxvec < minvec)
1073 if (WARN_ON_ONCE(dev->msi_enabled))
1076 nvec = pci_msi_vec_count(dev);
1087 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1092 rc = msi_capability_init(dev, nvec, affd);
1105 /* deprecated, don't use */
1106 int pci_enable_msi(struct pci_dev *dev)
1108 int rc = __pci_enable_msi_range(dev, 1, 1, NULL);
1113 EXPORT_SYMBOL(pci_enable_msi);
1115 static int __pci_enable_msix_range(struct pci_dev *dev,
1116 struct msix_entry *entries, int minvec,
1117 int maxvec, struct irq_affinity *affd,
1120 int rc, nvec = maxvec;
1122 if (maxvec < minvec)
1125 if (WARN_ON_ONCE(dev->msix_enabled))
1130 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1135 rc = __pci_enable_msix(dev, entries, nvec, affd, flags);
1149 * pci_enable_msix_range - configure device's MSI-X capability structure
1150 * @dev: pointer to the pci_dev data structure of MSI-X device function
1151 * @entries: pointer to an array of MSI-X entries
1152 * @minvec: minimum number of MSI-X IRQs requested
1153 * @maxvec: maximum number of MSI-X IRQs requested
1155 * Setup the MSI-X capability structure of device function with a maximum
1156 * possible number of interrupts in the range between @minvec and @maxvec
1157 * upon its software driver call to request for MSI-X mode enabled on its
1158 * hardware device function. It returns a negative errno if an error occurs.
1159 * If it succeeds, it returns the actual number of interrupts allocated and
1160 * indicates the successful configuration of MSI-X capability structure
1161 * with new allocated MSI-X interrupts.
1163 int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
1164 int minvec, int maxvec)
1166 return __pci_enable_msix_range(dev, entries, minvec, maxvec, NULL, 0);
1168 EXPORT_SYMBOL(pci_enable_msix_range);
1171 * pci_alloc_irq_vectors_affinity - allocate multiple IRQs for a device
1172 * @dev: PCI device to operate on
1173 * @min_vecs: minimum number of vectors required (must be >= 1)
1174 * @max_vecs: maximum (desired) number of vectors
1175 * @flags: flags or quirks for the allocation
1176 * @affd: optional description of the affinity requirements
1178 * Allocate up to @max_vecs interrupt vectors for @dev, using MSI-X or MSI
1179 * vectors if available, and fall back to a single legacy vector
1180 * if neither is available. Return the number of vectors allocated,
1181 * (which might be smaller than @max_vecs) if successful, or a negative
1182 * error code on error. If less than @min_vecs interrupt vectors are
1183 * available for @dev the function will fail with -ENOSPC.
1185 * To get the Linux IRQ number used for a vector that can be passed to
1186 * request_irq() use the pci_irq_vector() helper.
1188 int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
1189 unsigned int max_vecs, unsigned int flags,
1190 struct irq_affinity *affd)
1192 struct irq_affinity msi_default_affd = {0};
1193 int msix_vecs = -ENOSPC;
1194 int msi_vecs = -ENOSPC;
1196 if (flags & PCI_IRQ_AFFINITY) {
1198 affd = &msi_default_affd;
1204 if (flags & PCI_IRQ_MSIX) {
1205 msix_vecs = __pci_enable_msix_range(dev, NULL, min_vecs,
1206 max_vecs, affd, flags);
1211 if (flags & PCI_IRQ_MSI) {
1212 msi_vecs = __pci_enable_msi_range(dev, min_vecs, max_vecs,
1218 /* use legacy IRQ if allowed */
1219 if (flags & PCI_IRQ_LEGACY) {
1220 if (min_vecs == 1 && dev->irq) {
1222 * Invoke the affinity spreading logic to ensure that
1223 * the device driver can adjust queue configuration
1224 * for the single interrupt case.
1227 irq_create_affinity_masks(1, affd);
1233 if (msix_vecs == -ENOSPC)
1237 EXPORT_SYMBOL(pci_alloc_irq_vectors_affinity);
1240 * pci_free_irq_vectors - free previously allocated IRQs for a device
1241 * @dev: PCI device to operate on
1243 * Undoes the allocations and enabling in pci_alloc_irq_vectors().
1245 void pci_free_irq_vectors(struct pci_dev *dev)
1247 pci_disable_msix(dev);
1248 pci_disable_msi(dev);
1250 EXPORT_SYMBOL(pci_free_irq_vectors);
1253 * pci_irq_vector - return Linux IRQ number of a device vector
1254 * @dev: PCI device to operate on
1255 * @nr: device-relative interrupt vector index (0-based).
1257 int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
1259 if (dev->msix_enabled) {
1260 struct msi_desc *entry;
1263 for_each_pci_msi_entry(entry, dev) {
1272 if (dev->msi_enabled) {
1273 struct msi_desc *entry = first_pci_msi_entry(dev);
1275 if (WARN_ON_ONCE(nr >= entry->nvec_used))
1278 if (WARN_ON_ONCE(nr > 0))
1282 return dev->irq + nr;
1284 EXPORT_SYMBOL(pci_irq_vector);
1287 * pci_irq_get_affinity - return the affinity of a particular MSI vector
1288 * @dev: PCI device to operate on
1289 * @nr: device-relative interrupt vector index (0-based).
1291 const struct cpumask *pci_irq_get_affinity(struct pci_dev *dev, int nr)
1293 if (dev->msix_enabled) {
1294 struct msi_desc *entry;
1297 for_each_pci_msi_entry(entry, dev) {
1299 return &entry->affinity->mask;
1304 } else if (dev->msi_enabled) {
1305 struct msi_desc *entry = first_pci_msi_entry(dev);
1307 if (WARN_ON_ONCE(!entry || !entry->affinity ||
1308 nr >= entry->nvec_used))
1311 return &entry->affinity[nr].mask;
1313 return cpu_possible_mask;
1316 EXPORT_SYMBOL(pci_irq_get_affinity);
1319 * pci_irq_get_node - return the NUMA node of a particular MSI vector
1320 * @pdev: PCI device to operate on
1321 * @vec: device-relative interrupt vector index (0-based).
1323 int pci_irq_get_node(struct pci_dev *pdev, int vec)
1325 const struct cpumask *mask;
1327 mask = pci_irq_get_affinity(pdev, vec);
1329 return local_memory_node(cpu_to_node(cpumask_first(mask)));
1330 return dev_to_node(&pdev->dev);
1332 EXPORT_SYMBOL(pci_irq_get_node);
1334 struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
1336 return to_pci_dev(desc->dev);
1338 EXPORT_SYMBOL(msi_desc_to_pci_dev);
1340 void *msi_desc_to_pci_sysdata(struct msi_desc *desc)
1342 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1344 return dev->bus->sysdata;
1346 EXPORT_SYMBOL_GPL(msi_desc_to_pci_sysdata);
1348 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
1350 * pci_msi_domain_write_msg - Helper to write MSI message to PCI config space
1351 * @irq_data: Pointer to interrupt data of the MSI interrupt
1352 * @msg: Pointer to the message
1354 void pci_msi_domain_write_msg(struct irq_data *irq_data, struct msi_msg *msg)
1356 struct msi_desc *desc = irq_data_get_msi_desc(irq_data);
1359 * For MSI-X desc->irq is always equal to irq_data->irq. For
1360 * MSI only the first interrupt of MULTI MSI passes the test.
1362 if (desc->irq == irq_data->irq)
1363 __pci_write_msi_msg(desc, msg);
1367 * pci_msi_domain_calc_hwirq - Generate a unique ID for an MSI source
1368 * @dev: Pointer to the PCI device
1369 * @desc: Pointer to the MSI descriptor
1371 * The ID number is only used within the irqdomain.
1373 irq_hw_number_t pci_msi_domain_calc_hwirq(struct pci_dev *dev,
1374 struct msi_desc *desc)
1376 return (irq_hw_number_t)desc->msi_attrib.entry_nr |
1377 pci_dev_id(dev) << 11 |
1378 (pci_domain_nr(dev->bus) & 0xFFFFFFFF) << 27;
1381 static inline bool pci_msi_desc_is_multi_msi(struct msi_desc *desc)
1383 return !desc->msi_attrib.is_msix && desc->nvec_used > 1;
1387 * pci_msi_domain_check_cap - Verify that @domain supports the capabilities
1389 * @domain: The interrupt domain to check
1390 * @info: The domain info for verification
1391 * @dev: The device to check
1394 * 0 if the functionality is supported
1395 * 1 if Multi MSI is requested, but the domain does not support it
1396 * -ENOTSUPP otherwise
1398 int pci_msi_domain_check_cap(struct irq_domain *domain,
1399 struct msi_domain_info *info, struct device *dev)
1401 struct msi_desc *desc = first_pci_msi_entry(to_pci_dev(dev));
1403 /* Special handling to support __pci_enable_msi_range() */
1404 if (pci_msi_desc_is_multi_msi(desc) &&
1405 !(info->flags & MSI_FLAG_MULTI_PCI_MSI))
1407 else if (desc->msi_attrib.is_msix && !(info->flags & MSI_FLAG_PCI_MSIX))
1413 static int pci_msi_domain_handle_error(struct irq_domain *domain,
1414 struct msi_desc *desc, int error)
1416 /* Special handling to support __pci_enable_msi_range() */
1417 if (pci_msi_desc_is_multi_msi(desc) && error == -ENOSPC)
1423 #ifdef GENERIC_MSI_DOMAIN_OPS
1424 static void pci_msi_domain_set_desc(msi_alloc_info_t *arg,
1425 struct msi_desc *desc)
1428 arg->hwirq = pci_msi_domain_calc_hwirq(msi_desc_to_pci_dev(desc),
1432 #define pci_msi_domain_set_desc NULL
1435 static struct msi_domain_ops pci_msi_domain_ops_default = {
1436 .set_desc = pci_msi_domain_set_desc,
1437 .msi_check = pci_msi_domain_check_cap,
1438 .handle_error = pci_msi_domain_handle_error,
1441 static void pci_msi_domain_update_dom_ops(struct msi_domain_info *info)
1443 struct msi_domain_ops *ops = info->ops;
1446 info->ops = &pci_msi_domain_ops_default;
1448 if (ops->set_desc == NULL)
1449 ops->set_desc = pci_msi_domain_set_desc;
1450 if (ops->msi_check == NULL)
1451 ops->msi_check = pci_msi_domain_check_cap;
1452 if (ops->handle_error == NULL)
1453 ops->handle_error = pci_msi_domain_handle_error;
1457 static void pci_msi_domain_update_chip_ops(struct msi_domain_info *info)
1459 struct irq_chip *chip = info->chip;
1462 if (!chip->irq_write_msi_msg)
1463 chip->irq_write_msi_msg = pci_msi_domain_write_msg;
1464 if (!chip->irq_mask)
1465 chip->irq_mask = pci_msi_mask_irq;
1466 if (!chip->irq_unmask)
1467 chip->irq_unmask = pci_msi_unmask_irq;
1471 * pci_msi_create_irq_domain - Create a MSI interrupt domain
1472 * @fwnode: Optional fwnode of the interrupt controller
1473 * @info: MSI domain info
1474 * @parent: Parent irq domain
1476 * Updates the domain and chip ops and creates a MSI interrupt domain.
1479 * A domain pointer or NULL in case of failure.
1481 struct irq_domain *pci_msi_create_irq_domain(struct fwnode_handle *fwnode,
1482 struct msi_domain_info *info,
1483 struct irq_domain *parent)
1485 struct irq_domain *domain;
1487 if (WARN_ON(info->flags & MSI_FLAG_LEVEL_CAPABLE))
1488 info->flags &= ~MSI_FLAG_LEVEL_CAPABLE;
1490 if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
1491 pci_msi_domain_update_dom_ops(info);
1492 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
1493 pci_msi_domain_update_chip_ops(info);
1495 info->flags |= MSI_FLAG_ACTIVATE_EARLY;
1496 if (IS_ENABLED(CONFIG_GENERIC_IRQ_RESERVATION_MODE))
1497 info->flags |= MSI_FLAG_MUST_REACTIVATE;
1499 /* PCI-MSI is oneshot-safe */
1500 info->chip->flags |= IRQCHIP_ONESHOT_SAFE;
1502 domain = msi_create_irq_domain(fwnode, info, parent);
1506 irq_domain_update_bus_token(domain, DOMAIN_BUS_PCI_MSI);
1509 EXPORT_SYMBOL_GPL(pci_msi_create_irq_domain);
1512 * Users of the generic MSI infrastructure expect a device to have a single ID,
1513 * so with DMA aliases we have to pick the least-worst compromise. Devices with
1514 * DMA phantom functions tend to still emit MSIs from the real function number,
1515 * so we ignore those and only consider topological aliases where either the
1516 * alias device or RID appears on a different bus number. We also make the
1517 * reasonable assumption that bridges are walked in an upstream direction (so
1518 * the last one seen wins), and the much braver assumption that the most likely
1519 * case is that of PCI->PCIe so we should always use the alias RID. This echoes
1520 * the logic from intel_irq_remapping's set_msi_sid(), which presumably works
1521 * well enough in practice; in the face of the horrible PCIe<->PCI-X conditions
1522 * for taking ownership all we can really do is close our eyes and hope...
1524 static int get_msi_id_cb(struct pci_dev *pdev, u16 alias, void *data)
1527 u8 bus = PCI_BUS_NUM(*pa);
1529 if (pdev->bus->number != bus || PCI_BUS_NUM(alias) != bus)
1536 * pci_msi_domain_get_msi_rid - Get the MSI requester id (RID)
1537 * @domain: The interrupt domain
1538 * @pdev: The PCI device.
1540 * The RID for a device is formed from the alias, with a firmware
1541 * supplied mapping applied
1545 u32 pci_msi_domain_get_msi_rid(struct irq_domain *domain, struct pci_dev *pdev)
1547 struct device_node *of_node;
1548 u32 rid = pci_dev_id(pdev);
1550 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1552 of_node = irq_domain_get_of_node(domain);
1553 rid = of_node ? of_msi_map_rid(&pdev->dev, of_node, rid) :
1554 iort_msi_map_rid(&pdev->dev, rid);
1560 * pci_msi_get_device_domain - Get the MSI domain for a given PCI device
1561 * @pdev: The PCI device
1563 * Use the firmware data to find a device-specific MSI domain
1564 * (i.e. not one that is set as a default).
1566 * Returns: The corresponding MSI domain or NULL if none has been found.
1568 struct irq_domain *pci_msi_get_device_domain(struct pci_dev *pdev)
1570 struct irq_domain *dom;
1571 u32 rid = pci_dev_id(pdev);
1573 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1574 dom = of_msi_map_get_device_domain(&pdev->dev, rid);
1576 dom = iort_get_device_domain(&pdev->dev, rid);
1579 #endif /* CONFIG_PCI_MSI_IRQ_DOMAIN */